US20050240746A1 - Process and apparatus for memory mapping - Google Patents
Process and apparatus for memory mapping Download PDFInfo
- Publication number
- US20050240746A1 US20050240746A1 US10/830,739 US83073904A US2005240746A1 US 20050240746 A1 US20050240746 A1 US 20050240746A1 US 83073904 A US83073904 A US 83073904A US 2005240746 A1 US2005240746 A1 US 2005240746A1
- Authority
- US
- United States
- Prior art keywords
- memories
- class
- user
- basic
- mapping
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F12/00—Accessing, addressing or allocating within memory systems or architectures
- G06F12/02—Addressing or allocation; Relocation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
Definitions
- This invention relates to creation of IC floorplans, and particularly to mapping user-defined memories into standard or basic memories for implementation in basic IC platforms.
- Integrated circuits are used in a wide range electronic devices produced by a large number of device manufacturers. In practice, ICs are seldom manufactured (fabricated) by the electronic device manufacturer. Instead ICs are manufactured by an IC foundry to the specifications of the electronic device manufacturer. The design of the IC is usually the result of corroboration between the device manufacturer and the IC foundry.
- ASIC application-specific integrated circuit
- HDL hardware description language
- RTL synthesizable register transfer language
- IC foundries have developed standard, or base, platforms containing silicon layers of an IC, but without metal interconnection layers.
- the silicon layers are configured into gates that can be configured into cells using tools supplied by the IC foundry.
- the chip designer designs additional metal layers for the base platform to thereby configure the chip into a custom ASIC employing the customer's intellectual property.
- the IC foundry ordinarily supplies tools to the IC designer to enable the designer to quickly and accurately configure the base platform to a custom ASIC compatible with the foundry's fabrication technology.
- An example of such a configurable base platform is the RapidChip® platform available from LSI Logic Corporation of Milpitas, California.
- the RapidChip platform permits the development of complex, high-density ASICs in minimal time with significantly reduced design and manufacturing risks and costs.
- the design effort can be considered as encompassing several stages. After the chip size and the placement of the I/O cells has been selected, the megacells, including memories, are placed. Thereafter, standard cells are place to complete the chip.
- the present invention deals with placement of memories, and particularly to mapping user's custom memories to standard or basic memories that are incorporated into the base platform.
- the design created by the IC designer may contain user-defined memories that are different from the basic memories.
- the present invention is directed to techniques for mapping user-defined memories to basic memories so that the user-defined memories can be implemented in basic memories on the base platform.
- a plurality of user-defined memories are mapped to pre-defined basic memories by dividing the user-defined memories into classes of similar user-defined memories, which in some embodiments are memories with the same type and capacity.
- a maximal ratio (max i,j (USED i,j /AVAIL i,j )) is calculated of basic memories that have been mapped (USED i,j ) to basic memories that are available for mapping (AVAIL i,j ). Mapping techniques are selected for members of each class of user-defined memories that minimizes the ratio. The calculated maximal ratio and selection of the mapping technique are iteratively repeated for each class.
- an estimate is calculated of the number of different memory mappings necessary to map the user-defined memories to the basic memories. If the number of different memory mappings exceeds a threshold, a mapping price is calculated for each class. Queues of similar memories are created for each memory type, with the memories arranged in order of mapping prices. A mapping technique is selected for one memory of a class, and is applied to all of the memories of the class. The process is repeated for each class of each queue.
- a computer useable medium contains a computer readable program comprising computer readable program code for addressing data to cause a computer to carry out the process of the invention.
- Memory mapping can be considered in two parts: (1) whether it is even possible to map a given set of user-defined memories into a predetermined set of basic memories, and (2) if it is possible, to find the optimal solution of such a mapping.
- each memory M may be defined by three parameters:
- Memory M can be mapped into the set of basic memories M 1 , M 2 , . . . , M K if memory M can be presented as the netlist that consists of memories M 1 , M 2 . . . . , M K and some logic cells.
- memory M having a capacity (k*C)
- k*C capacity
- memory M 1 can be mapped into a similar set of basic memories as memory M 2 .
- each basic memory has a capacity corresponding to one of the capacities of the set ⁇ C 1 , C 2 , C 3 , . . . ⁇ where C 1 ⁇ C 2 ⁇ C 3 ⁇ . . . .
- the user-defined memories are designated Mem 1 , Mem 2 , . . . , Mem N , where N is the number of user-defined memories.
- the goal is to map the user-defined memories into the given set of basic memories, namely to map the user-defined memories Mem 1 , Mem 2 , . . . , Mem N into basic memories M 1 , M 2 , . . . , M K .
- the number r k,j is the minimal possible number of basic memories with type t(Mem k ), width W and capacity C j into which the user memory Mem k can be mapped.
- k 1,2, . . . ,N, r k,1 ⁇ r k,2 ⁇ r k,3 .
- the number AVAIL i,j is the total number of available basic memories M basic with type t(M basic ) ⁇ i and capacity c(M basic ) ⁇ C j .
- BISR build-in self-repair
- the problem can be reduced to an integer programming of an NP-hard problem.
- a solution to an NP-hard problem requires an exponentially increasing computation time as the number of inputs linearly increases.
- a given design may contain hundreds or even thousands of user-defined memories (N is in the hundreds or thousands). Therefore, to minimize processing time and find a good solution to memory mapping, the present invention solves the memory mapping for classes of similar user-defined memories. Using this technique, the time to execute the process is O(N*log N), resulting in a solution that is close to optimal.
- two similar memories may have different width W and/or capacity C, such memories are not different for purposes of memory mapping according to the present invention.
- the user-defined memories are divided into the classes, with each class containing similar memories.
- L is the number of classes and CLASS 1 , CLASS 2 , . . . , CLASS L are the classes of similar user-defined memories.
- the user design will contain no more than about 10 memory classes, a number that is significantly less than the number N of user-defined memories.
- t(CLASS q ) is the class type (namely, the type of user-defined memories included in this class)
- MEM_MAP — NUM q ( cl — sz q +2)*( cl — sz q +1)/2
- the number MEM_MAP_NUM provides an estimate of the time required to search for the optimal mapping by examining all possible memory mappings. Since the number L of user-defined memory classes is not large compared to the number N of user-defined memories, the number MEM_MAP_NUM is not large. Consequently, the mapping technique can efficiently examine all the memory mappings. In practice, we have found that the mapping algorithm may be effectively applied to all mappings if MEM_MAP_NUM ⁇ 10 6 or 10 8 , the bound depending on the power of computer system processing the memory mapping.
- FIGURE is a flowchart of an embodiment of the process of the present invention. It is preferred that the process be carried out in a computer or data processor under the control of a computer readable program containing program code recorded on a memory medium, such as a recording disk of a disk drive.
- the value of cl_sz q defines the total number of user-defined memories to be mapped.
- the values cl_p q,j increase, whereas the value of cl_sz q does not change.
- the user-defined memories Mem k are divided into classes CLASS q of similar memory types t(CLASS q ).
- step 14 If at step 14 the value of MEM_MAP_NUM is smaller than some predetermined threshold (e.g., 10 6 or 10 8 , depending on computer processing power, then at step 16 the number of user-defined memories, cl_p q,j , is selected that minimizes the value of max i,j (USED i,j /AVAIL i,j ).
- the best memory mapping technique p k is selected for each user-defined memory, based on the current value of cl_p q,j .
- the mapping technique p k is output at step 22 .
- step 16 If at step 16 , the best mapping technique is not found, the process is aborted because there is not enough basic memory for mapping the user-defined memories.
- the capacity of CLASS q designated c(CLASS q )
- the width of CLASS q designated w(CLASS q )
- Price q,j 1.0 ⁇ c (CLASS q )* w (CLASS q )/ W/C j /cl — r q,i .
- Price q,j may be considered as the price that is paid for mapping the user-defined memory of class CLASS q into basic memories with capacity C j .
- values of price are:
- the process concludes at steps 20 and 22 described above.
- three queues of are created.
- the classes are sorted into ascending order of the value of price q,j , with each class being composed of user-defined memories of similar memory types.
- the three queues are (with the price given in parentheses):
- the first class of each queue is selected for evaluation.
- CLASS 1 (0.05) is selected from QUEUE 1
- CLASS 3 (0.29) is selected from QUEUE 3
- CLASS 2 (0.53) is selected from QUEUE 2 (the selection of CLASS 2 (0.53) is only because it was randomly placed as the first class of the queue).
- the classes in each queue are processed in linear order of the queue according to price.
- the first classes in the queues are selected and processed, then the next classes. In the event that the current class is the last class in the queue, an attempt to select the next class fails and the queue is completely examined.
- the currently selected class CLASS q in queue QUEUE j is completely mapped or if CUR_AVAIL j is smaller than the minimum number of available basic memories (CUR_AVAIL j ⁇ cl_r q,j ), then the next class in queue QUEUE j is selected. If QUEUE j has been completely examined, the process loops back to step 32 to select the next queue. If QUEUE j has not been completely examined, another CLASS q in QUEUE j is selected until a class in QUEUE j is found that has not been completely mapped.
- step 32 The process then loops to step 32 to supply the current value of cl_p q,j (which had been incremented at step 38 ) to step 20 .
- the present invention thus provides a technique and computer program that maps user-defined memories to basic memories, such as on a base platform, for creating an ASIC.
- the user-defined memories are divided into classes of memories having the same type and capacity. If, at step 16 , the number of different memory mappings is smaller than a threshold, such as 10 8 , a mapping technique is selected to map each user-defined memory to one or more basic memories. If, at step 24 , the number of different memory mappings is greater than the threshold, the groups are arranged in queues of single memory types and in an order based on the mapping price. A mapping technique is selected for each group of memories and is applied to the members of each group to map the user-defined memories to basic memories.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Description
- This invention relates to creation of IC floorplans, and particularly to mapping user-defined memories into standard or basic memories for implementation in basic IC platforms.
- Integrated circuits are used in a wide range electronic devices produced by a large number of device manufacturers. In practice, ICs are seldom manufactured (fabricated) by the electronic device manufacturer. Instead ICs are manufactured by an IC foundry to the specifications of the electronic device manufacturer. The design of the IC is usually the result of corroboration between the device manufacturer and the IC foundry.
- The design and manufacture of an application-specific IC, or ASIC, is usually a long and tedious process, requiring development of a hardware description language (HDL) description of the circuit, usually in a synthesizable register transfer language (synthesizable RTL), synthesizing the RTL description to a technology library of components, and ultimately fabricating the circuit into an IC chip. During the process, testing and re-designing is necessary to optimize cell placement to meet physical constrains, wire routing and timing constraints. The process is time consuming and costly.
- To reduce the time and cost of development of ASICs, IC foundries have developed standard, or base, platforms containing silicon layers of an IC, but without metal interconnection layers. The silicon layers are configured into gates that can be configured into cells using tools supplied by the IC foundry. The chip designer designs additional metal layers for the base platform to thereby configure the chip into a custom ASIC employing the customer's intellectual property. The IC foundry ordinarily supplies tools to the IC designer to enable the designer to quickly and accurately configure the base platform to a custom ASIC compatible with the foundry's fabrication technology. An example of such a configurable base platform is the RapidChip® platform available from LSI Logic Corporation of Milpitas, California. The RapidChip platform permits the development of complex, high-density ASICs in minimal time with significantly reduced design and manufacturing risks and costs.
- The design effort can be considered as encompassing several stages. After the chip size and the placement of the I/O cells has been selected, the megacells, including memories, are placed. Thereafter, standard cells are place to complete the chip. The present invention deals with placement of memories, and particularly to mapping user's custom memories to standard or basic memories that are incorporated into the base platform.
- Consider a base platform containing a predetermined number of basic memories of predetermined types. The design created by the IC designer may contain user-defined memories that are different from the basic memories. The present invention is directed to techniques for mapping user-defined memories to basic memories so that the user-defined memories can be implemented in basic memories on the base platform.
- In one embodiment of the present invention, a plurality of user-defined memories are mapped to pre-defined basic memories by dividing the user-defined memories into classes of similar user-defined memories, which in some embodiments are memories with the same type and capacity. A maximal ratio (maxi,j(USEDi,j/AVAILi,j)) is calculated of basic memories that have been mapped (USEDi,j) to basic memories that are available for mapping (AVAILi,j). Mapping techniques are selected for members of each class of user-defined memories that minimizes the ratio. The calculated maximal ratio and selection of the mapping technique are iteratively repeated for each class.
- In some embodiments an estimate is calculated of the number of different memory mappings necessary to map the user-defined memories to the basic memories. If the number of different memory mappings exceeds a threshold, a mapping price is calculated for each class. Queues of similar memories are created for each memory type, with the memories arranged in order of mapping prices. A mapping technique is selected for one memory of a class, and is applied to all of the memories of the class. The process is repeated for each class of each queue.
- If the number of different memory mappings does not exceed the threshold, all user-defined memories are mapped to basic memories.
- In other embodiments, a computer useable medium contains a computer readable program comprising computer readable program code for addressing data to cause a computer to carry out the process of the invention.
- The sole FIGURE is a flowchart of the process of carrying out the process of mapping memories according to an embodiment of the present invention
- Memory mapping can be considered in two parts: (1) whether it is even possible to map a given set of user-defined memories into a predetermined set of basic memories, and (2) if it is possible, to find the optimal solution of such a mapping.
- For purposes of the present invention, each memory M may be defined by three parameters:
-
- 1) capacity c(M) is the number of words that can be stored in memory M,
- 2) width w(M) is the bit size of words stored in memory M, and
- 3) type t(M).
The type t(M) has a value of 1, 2 or 3, where - t(M)=1 is assigned to memories that execute only one read or one write operation during each clock cycle, 1RW,
- t(M)=2 is assigned to memories that execute one read and one write operation during each clock cycle, 1R1W, or
- t(M)=3 is assigned to memories that execute not more than two read or write operations during each clock cycle, 2RW.
- Consider a user-defined memory M. Memory M can be mapped into the set of basic memories M1, M2, . . . , MK if memory M can be presented as the netlist that consists of memories M1, M2 . . . . , MK and some logic cells. Thus, memory M, having a capacity (k*C), can be mapped into the set of memories M1, M2, . . . , MK having the capacity C and the same width and type. If two user-defined memories M1 and M2 are such that t(M1)≦t(M2), w(M1)≦w(M2), c(M1)≦c(M2), then memory M1 can be mapped into a similar set of basic memories as memory M2.
- Consider a chip platform containing a predetermined number of basic memories. All of the basic memories have the same width W, and each basic memory has a capacity corresponding to one of the capacities of the set {C1, C2, C3, . . . } where C1<C2<C3< . . . . For ease of explanation, it will be assumed that the set of capacities is composed of three capacities Cj, where j=1,2,3, but the number of capacities (j) may be any convenient number. BASICi,j (i=1,2,3; j=1,2,3) is the number of the basic memories of type i and capacity Cj in the chip platform.
- The user-defined memories are designated Mem1, Mem2, . . . , MemN, where N is the number of user-defined memories. The goal is to map the user-defined memories into the given set of basic memories, namely to map the user-defined memories Mem1, Mem2, . . . , MemN into basic memories M1, M2, . . . , MK.
- For each j=1,2,3 and each k=1,2, . . . ,N, the number rk,j is the minimal possible number of basic memories with type t(Memk), width W and capacity Cj into which the user memory Memk can be mapped. For each k=1,2, . . . ,N, rk,1≦rk,2≧rk,3. This means that the number of basic memories required to map user-defined memory Memk increases as each basic memory capacity decreases. Where j∈{1,2,3}, there are three ways that each user-defined memory Memk (k=1,2, . . . ,N) can be mapped to the basic memory:
-
- 1) map memory Memk into rk,1 memories of capacity C1;
- 2) map memory Memk into rk,2 memories of capacity C2;
- 3) map memory Memk into rk,3 memories of capacity C3.
- The value pk (k=1,2, . . . ,N) is a number assigned to one of the ways of mapping the user memory Memk to a basic memory. Hence, the number pk designates a mapping technique. Where j∈{1,2,3}, pk∈{1,2,3}.
- In order to take into account the fact that the basic memory with type i1 and capacity Cj1 can be mapped into the basic memory with type i2 and capacity Cj2, for each i1≦i2 and j1≦j2 consider the numbers AVAILi,j (i=1,2,3; j=1,2,3) instead of the numbers BASICi,j, where
- The number AVAILi,j is the total number of available basic memories Mbasic with type t(Mbasic)≧i and capacity c(Mbasic)≧Cj.
- For each i=1,2,3, j=1,2,3, the number USEDi,j denotes the number of basic memories, Mbasic, of type t(Mbasic)≧i and capacity c(Mbasic)≧Cj into which the user-defined memories are already mapped by the mapping technique pk (k=1,2, . . . ,N).
- Using these definitions, the problem of memory mapping is defined as finding the mapping technique pk (k=1,2, . . . ,N) for the given user memories Memk (k=1,2, . . . ,N) and the given values BASICi,j (i=1,2,3, j=1,2,3, . . . ) and rk,j(k=1,2, . . . ,N, j=1,2,3, . . . ) such that for each i=1,2,3 and each j=1,2,3, USEDi,j is not greater than AVAILi,j, i.e., USEDi,j≦AVAILi,j. Where j∈{1,2,3}, these nine inequalities mean that the set of basic memories is enough for the mapping user memories Memk (k=1,2, . . . ,N).
- To obtain the best BISR (build-in self-repair) of memories, it is desirable to create optimal memory mapping. To accomplish this, it is important to use basic memories of different types as uniformly as practical. Therefore, a second part of the memory mapping problem is to find the mapping technique pk (k=1,2, . . . ,N) that minimizes the magnitude maxi,j(USEDi,j/AVAILi,j).
- The problem can be reduced to an integer programming of an NP-hard problem. However, a solution to an NP-hard problem requires an exponentially increasing computation time as the number of inputs linearly increases. In the present case, a given design may contain hundreds or even thousands of user-defined memories (N is in the hundreds or thousands). Therefore, to minimize processing time and find a good solution to memory mapping, the present invention solves the memory mapping for classes of similar user-defined memories. Using this technique, the time to execute the process is O(N*log N), resulting in a solution that is close to optimal.
- For purposes of the present invention two user-defined memories Mema and Memb are similar if ra,j=rb,j for each j=1,2,3 and these memories are the same type (t(Mema)=t(Memb)). Although two similar memories may have different width W and/or capacity C, such memories are not different for purposes of memory mapping according to the present invention.
- In accordance with the present invention, the user-defined memories are divided into the classes, with each class containing similar memories. L is the number of classes and CLASS1, CLASS2, . . . , CLASSL are the classes of similar user-defined memories. In most cases the user design will contain no more than about 10 memory classes, a number that is significantly less than the number N of user-defined memories.
- For each class CLASSq (q=1,2, . . . ,L), t(CLASSq) is the class type (namely, the type of user-defined memories included in this class), and cl_rq,j is the minimum number of basic memories of capacity Cj that is required for mapping any memory Memk from class CLASSq. Since Memk∈CLASSq, rk,j=cl_rq,j. For each q=1,2, . . . ,L, cl_pq,j is the number of user-defined memories in class CLASSq that have been mapped to the basic memories with capacity Cj (j=1,2,3). Thus,
- The number of different choices of values for cl_pq,j for CLASSq is
MEM — MAP — NUM q=(cl — sz q+2)*(cl — sz q+1)/2,
where cl_szq is the total number of user-defined memories of class CLASSq being mapped. Consequently, the total number MEM_MAP_NUM of different memory mappings can be obtained as follows: - The number MEM_MAP_NUM provides an estimate of the time required to search for the optimal mapping by examining all possible memory mappings. Since the number L of user-defined memory classes is not large compared to the number N of user-defined memories, the number MEM_MAP_NUM is not large. Consequently, the mapping technique can efficiently examine all the memory mappings. In practice, we have found that the mapping algorithm may be effectively applied to all mappings if MEM_MAP_NUM≦106 or 108, the bound depending on the power of computer system processing the memory mapping.
- The sole FIGURE is a flowchart of an embodiment of the process of the present invention. It is preferred that the process be carried out in a computer or data processor under the control of a computer readable program containing program code recorded on a memory medium, such as a recording disk of a disk drive.
- The process starts with the empty memory mapping cl_pq,j=0 (q=1,2, . . . ,L, j=1,2,3), meaning that no user memory has yet been mapped into the basic memories. The value of cl_szq defines the total number of user-defined memories to be mapped. During the execution of the process the values cl_pq,j increase, whereas the value of cl_szq does not change. A class CLASSq is completely mapped if all the user memories of this class are mapped into the basic memories, that is Σj=1,2,3 (cl_Pq,j)=cl_szq.
- At
step 10, the definitions of the user-defined memories Mem1, Mem2, . . . , MemN are input to the process, as are value of BASICi,j, which is the number of basic memories of type i and capacity Cj in the chip platform, and value of rk,j, which is the minimum possible number of basic memories with type t(Memk), width W and capacity Cj such that the user-defined memories Memk can mapped to the basic memories, where i=1,2,3 . . . , j is a number (j=1,2,3 . . . ) identifying a standard capacity Cj and k=1,2, . . . , N. - At
step 12, the total number of available basic memories AVAILi,j is calculated for each i and j as
Atstep 14, the user-defined memories Memk are divided into classes CLASSq of similar memory types t(CLASSq). As described above, user-defined memories Ma and Mb are similar if they are of the same type (t(Mema)=t(Memb)) and ra,j=rb,j. The number, cl_rq,j, of user-defined memories in each class CLASSq is identified, and the number, cl_szq, of user memories in the class CLASSq is identified. The total number, MEM_MAP_NUM, of different memory mappings is calculated as - If at
step 14 the value of MEM_MAP_NUM is smaller than some predetermined threshold (e.g., 106 or 108, depending on computer processing power, then atstep 16 the number of user-defined memories, cl_pq,j, is selected that minimizes the value of maxi,j(USEDi,j/AVAILi,j). Atstep 20 the best memory mapping technique pk is selected for each user-defined memory, based on the current value of cl_pq,j. The mapping technique pk is output atstep 22. - If at
step 16, the best mapping technique is not found, the process is aborted because there is not enough basic memory for mapping the user-defined memories. - If at
step 14, the value of MEM_MAP_NUM is greater than the predetermined threshold, then atstep 24, values of the capacity of CLASSq, designated c(CLASSq), the width of CLASSq, designated w(CLASSq) and a penalty, designated priceq,j, are calculated. More particularly, for each q=1,2, . . . ,L, the capacity, c(CLASSq), and width, w(CLASSq), of this class is equal to the capacity c(Memk) and the width w(Memk) of some user-defined memory Memk of this class. (There is a small degree of inaccuracy because different memories Memk of class CLASSq may have different capacities and widths, but this inaccuracy is not material to the result.) - For each q=1,2, . . . ,L and each j=1,2,3, the price is calculated as
priceq,j=1.0−c(CLASSq)*w(CLASSq)/W/C j /cl — r q,i.
Priceq,j may be considered as the price that is paid for mapping the user-defined memory of class CLASSq into basic memories with capacity Cj. - Consider the example set forth in Table I that lists of user memories of the user design, N=168, W=20, C1=512, C2=1024, C3=2048.
TABLE I user memories type Width Capacity Memk, k t (Memk) w (Memk) c (Memk) rk,1 rk,2 rk,3 1-20 1 (1RW) 10 78 1 1 1 21-40 1 (1RW) 20 76 1 1 1 41-42 3 (2RW) 10 2919 5 3 2 43-62 1 (1RW) 10 58 1 1 1 63-66 2 (1R1W) 10 1946 3 2 1 67-168 1 (1RW) 10 973 1 1 1 - The user-defined memories in
lines 1, 2, 4 and 6 of Table I are similar and can be included in a single class. Table II presents the list of user memory classes.TABLE II user Class memor- Type CLASSq, ies t q Memk, k (CLASSq) cl_szq cl_rq,1 cl_rq,2 cl_rq,3 1 1-40, 1 (1RW) 162 1 1 1 43-62, 67-168 2 63-66 2 (1R1W) 4 3 2 1 3 41-42 3 (2RW) 2 5 3 2 - From the above example, c(CLASS1) and w(CLASS1) can be defined from Table I with four different sets of values (see
lines 1, 2, 4 and 6): 78 and 10, 76 and 20, 58 and 10, 973 and 10. Choosing the fourth line, c(CLASS1)=973 and w(CLASS1)=10. Calculating the price for q=3 and j=1, from Table I it can be seen that c(CLASS3) is 2919, w(CLASS3) is 10 and cl_r3,1 is 5 and from the parameters of the user memory, W is 20 and C1 is 512. Therefore, price3,1=1.0−c(CLASS3)*w(CLASS3)/W/C1/cl_r3,1=1.0−2919*10/20/512/5=0.43. Thus, in the example, values of price are: -
- price3,1=1.0−2919*10/20/512/5=0.43.
- price3,2=1.0−2919*10/20/1024/3=0.53.
- price3,3=1.0−2919*10/20/2048/2=0.29.
- price2,1=0.37.
- price2,2=0.53.
- price2,3=0.53.
- price1,1=0.05.
- price1,2=0.53.
- price1,3=0.76.
- MEM_MAP_NUM can be calculated from Table II as
If the predetermined threshold is 108, 1202940<108 so atstep 16 the algorithm may be applied to examine all possible memory mappings. - If MEM_MAP_NUM is greater than the threshold, then the classes CLASSq are organized into queues of same-type memories and one memory from each class is mapped. The mapping technique applied to the one memory is repeated for each other memory of the class. As noted above, at step 24 c(CLASSq), w(CLASSq) and priceq,j are calculated. At
step 26, the value of cl_Pq,j is set to zero (q=1,2, . . . ,L, j=1,2,3), meaning that no user memory is mapped into the basic memories. -
Step 28 is an iterative repeating ofsteps step 18, i=0 and all memory types have been examined. The process concludes atsteps - If at step 28 i≠0, the process continues to step 30 where a plurality of queues are created, QUEUEj (j=1,2,3, . . . ) for the classes of type i only. In the case of j∈{1,2,3}, three queues of are created. The classes are sorted into ascending order of the value of priceq,j, with each class being composed of user-defined memories of similar memory types. In the above example the three queues are (with the price given in parentheses):
-
- QUEUE1: CLASS1(0.05), CLASS2(0.37), CLASS3(0.43).
- QUEUE2: CLASS2(0.53), CLASS1(0.53), CLASS3(0.53) (random order because all the prices are equal).
- QUEUE3: CLASS3(0.29), CLASS2(0.53), CLASS3(0.76).
- The first class of each queue is selected for evaluation. In the example, CLASS1(0.05) is selected from QUEUE1, CLASS3(0.29) is selected from QUEUE3, and CLASS2(0.53) is selected from QUEUE2 (the selection of CLASS2(0.53) is only because it was randomly placed as the first class of the queue). For each selected class, the value of USED is calculated based on current values of cl_pq,j (j=1,2,3, . . . ). The value of CUR_AVAIL is calculated as min1<=x<=j(AVAILi,x−USEDi,x) (j=1,2,3).
- The classes in each queue are processed in linear order of the queue according to price. The first classes in the queues are selected and processed, then the next classes. In the event that the current class is the last class in the queue, an attempt to select the next class fails and the queue is completely examined.
- At
step 32, a set J ofindexes 1≦j≦3, where j∈{1,2,3}∈J, is created such that CUR_AVAILj>0 and the queue QUEUEj is not completely examined. If J is an empty set, and if all classes of type i are completely mapped out, i is decremented, i=i−1. If i≠0, the process ofsteps step 32, J is an empty set and type i is not completely mapped out, then the process is aborted because there is not enough basic memory for mapping the user-defined memories. - At
step 34, if J is not an empty set, an index j is selected from set J (j∈J) such that CUR_AVAILj/AVAILi,j=maxx∈J(CUR_AVAILj/AVAILi,x). Atstep 36, if the currently selected class CLASSq in queue QUEUEj is completely mapped or if CUR_AVAILj is smaller than the minimum number of available basic memories (CUR_AVAILj<cl_rq,j), then the next class in queue QUEUEj is selected. If QUEUEj has been completely examined, the process loops back to step 32 to select the next queue. If QUEUEj has not been completely examined, another CLASSq in QUEUEj is selected until a class in QUEUEj is found that has not been completely mapped. - If the currently selected class CLASSq in queue QUEUEj is not completely mapped and CUR_AVAILj≧cl_rq,j, then at
step 38 the value of the mapping technique number, cl_pq,j, is incremented by 1 (cl_pq,j=cl_pq,j+1), and either -
- if 1≦x≦j, CUR_AVAILx is assigned equal to CUR_AVAILx−cl_rq,j, or
- if j<x≦3, CUR_AVAILx is assigned equal to min(CUR_AVAILx, CUR_AVAILj).
- The process then loops to step 32 to supply the current value of cl_pq,j (which had been incremented at step 38) to step 20.
- The present invention thus provides a technique and computer program that maps user-defined memories to basic memories, such as on a base platform, for creating an ASIC. The user-defined memories are divided into classes of memories having the same type and capacity. If, at
step 16, the number of different memory mappings is smaller than a threshold, such as 108, a mapping technique is selected to map each user-defined memory to one or more basic memories. If, atstep 24, the number of different memory mappings is greater than the threshold, the groups are arranged in queues of single memory types and in an order based on the mapping price. A mapping technique is selected for each group of memories and is applied to the members of each group to map the user-defined memories to basic memories. - Although the present invention has been described with reference to preferred embodiments, workers skilled in the art will recognize that changes may be made in form and detail without departing from the spirit and scope of the invention.
Claims (20)
priceq,j=1.0−c(CLASSq)*w(CLASSq)/W/C j /cl — r q,i,
priceq,j=1.0−c(CLASSq)*w(CLASSq)/W/C j /cl — r q,i,
priceq,j=1.0−c(CLASSq)*w(CLASSq)/W/C j /cl — r q,i,
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US10/830,739 US7219321B2 (en) | 2004-04-23 | 2004-04-23 | Process and apparatus for memory mapping |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US10/830,739 US7219321B2 (en) | 2004-04-23 | 2004-04-23 | Process and apparatus for memory mapping |
Publications (2)
Publication Number | Publication Date |
---|---|
US20050240746A1 true US20050240746A1 (en) | 2005-10-27 |
US7219321B2 US7219321B2 (en) | 2007-05-15 |
Family
ID=35137822
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US10/830,739 Active 2025-06-19 US7219321B2 (en) | 2004-04-23 | 2004-04-23 | Process and apparatus for memory mapping |
Country Status (1)
Country | Link |
---|---|
US (1) | US7219321B2 (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060010092A1 (en) * | 2004-06-23 | 2006-01-12 | Lsi Logic Corporation | Yield driven memory placement system |
US7219321B2 (en) | 2004-04-23 | 2007-05-15 | Lsi Logic Corporation | Process and apparatus for memory mapping |
US7315993B2 (en) * | 2004-11-30 | 2008-01-01 | Lsi Logic Corporation | Verification of RRAM tiling netlist |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7313775B2 (en) * | 2005-04-06 | 2007-12-25 | Lsi Corporation | Integrated circuit with relocatable processor hardmac |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4400794A (en) * | 1981-11-17 | 1983-08-23 | Burroughs Corporation | Memory mapping unit |
US6601205B1 (en) * | 2000-09-29 | 2003-07-29 | Infineon Technologies Ag | Method to descramble the data mapping in memory circuits |
US6611952B1 (en) * | 2000-12-21 | 2003-08-26 | Shiv Prakash | Interactive memory allocation in a behavioral synthesis tool |
US20040193829A1 (en) * | 2001-07-30 | 2004-09-30 | Woo Steven C. | Consolidation of allocated memory to reduce power consumption |
US6871328B1 (en) * | 2002-11-14 | 2005-03-22 | Altera Corporation | Method for mapping logic design memory into physical memory device of a programmable logic device |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7219321B2 (en) | 2004-04-23 | 2007-05-15 | Lsi Logic Corporation | Process and apparatus for memory mapping |
-
2004
- 2004-04-23 US US10/830,739 patent/US7219321B2/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4400794A (en) * | 1981-11-17 | 1983-08-23 | Burroughs Corporation | Memory mapping unit |
US6601205B1 (en) * | 2000-09-29 | 2003-07-29 | Infineon Technologies Ag | Method to descramble the data mapping in memory circuits |
US6611952B1 (en) * | 2000-12-21 | 2003-08-26 | Shiv Prakash | Interactive memory allocation in a behavioral synthesis tool |
US20040193829A1 (en) * | 2001-07-30 | 2004-09-30 | Woo Steven C. | Consolidation of allocated memory to reduce power consumption |
US6871328B1 (en) * | 2002-11-14 | 2005-03-22 | Altera Corporation | Method for mapping logic design memory into physical memory device of a programmable logic device |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7219321B2 (en) | 2004-04-23 | 2007-05-15 | Lsi Logic Corporation | Process and apparatus for memory mapping |
US20060010092A1 (en) * | 2004-06-23 | 2006-01-12 | Lsi Logic Corporation | Yield driven memory placement system |
US7168052B2 (en) * | 2004-06-23 | 2007-01-23 | Lsi Logic Corporation | Yield driven memory placement system |
US7315993B2 (en) * | 2004-11-30 | 2008-01-01 | Lsi Logic Corporation | Verification of RRAM tiling netlist |
Also Published As
Publication number | Publication date |
---|---|
US7219321B2 (en) | 2007-05-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US6516456B1 (en) | Method and apparatus for selectively viewing nets within a database editor tool | |
US6240542B1 (en) | Poly routing for chip interconnects with minimal impact on chip performance | |
US9251306B2 (en) | Alignment net insertion for straightening the datapath in a force-directed placer | |
CN101187957A (en) | System and method for designing multiple latch unit layout of integrated circuit public clock domain clock aware placement | |
US6990651B2 (en) | Advanced design format library for integrated circuit design synthesis and floorplanning tools | |
US20050268267A1 (en) | Methods and systems for mixed-mode physical synthesis in electronic design automation | |
Sarrafzadeh et al. | Modern placement techniques | |
US6948143B2 (en) | Constrained optimization with linear constraints to remove overlap among cells of an integrated circuit | |
US7013441B2 (en) | Method for modeling integrated circuit yield | |
US6564361B1 (en) | Method and apparatus for timing driven resynthesis | |
US6543032B1 (en) | Method and apparatus for local resynthesis of logic trees with multiple cost functions | |
US9390211B1 (en) | Circuit placement based on fuzzy clustering | |
US8954838B2 (en) | Presenting data in a tabular format | |
US7219321B2 (en) | Process and apparatus for memory mapping | |
US6681373B1 (en) | Method and apparatus for dynamic buffer and inverter tree optimization | |
US7430699B2 (en) | Trading propensity-based clustering of circuit elements in a circuit design | |
US6532582B1 (en) | Method and apparatus for optimal critical netlist area selection | |
US9886539B2 (en) | Designing circuits using pseudohierarchy | |
US6634014B1 (en) | Delay/load estimation for use in integrated circuit design | |
US7302377B1 (en) | Accelerated event queue for logic simulation | |
US20090164184A1 (en) | Method and System for Implementing a Complex System or Process | |
US6637011B1 (en) | Method and apparatus for quick search for identities applicable to specified formula | |
US7103858B2 (en) | Process and apparatus for characterizing intellectual property for integration into an IC platform environment | |
US10325050B2 (en) | User-defined partitions for logical and physical circuit syntheses | |
US20070044056A1 (en) | Macro block placement by pin connectivity |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AS | Assignment |
Owner name: LSI LOGIC CORPORATION, CALIFORNIA Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:NIKITIN, ANDREY A.;ANDREEV, ALEXANDER E.;BOLOTOV, ANATOLI A.;REEL/FRAME:015265/0988;SIGNING DATES FROM 20040330 TO 20040407 |
|
STCF | Information on status: patent grant |
Free format text: PATENTED CASE |
|
CC | Certificate of correction | ||
FEPP | Fee payment procedure |
Free format text: PAYOR NUMBER ASSIGNED (ORIGINAL EVENT CODE: ASPN); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY |
|
FPAY | Fee payment |
Year of fee payment: 4 |
|
AS | Assignment |
Owner name: DEUTSCHE BANK AG NEW YORK BRANCH, AS COLLATERAL AG Free format text: PATENT SECURITY AGREEMENT;ASSIGNORS:LSI CORPORATION;AGERE SYSTEMS LLC;REEL/FRAME:032856/0031 Effective date: 20140506 |
|
AS | Assignment |
Owner name: LSI CORPORATION, CALIFORNIA Free format text: CHANGE OF NAME;ASSIGNOR:LSI LOGIC CORPORATION;REEL/FRAME:033102/0270 Effective date: 20070406 |
|
FPAY | Fee payment |
Year of fee payment: 8 |
|
AS | Assignment |
Owner name: AVAGO TECHNOLOGIES GENERAL IP (SINGAPORE) PTE. LTD Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:LSI CORPORATION;REEL/FRAME:035390/0388 Effective date: 20140814 |
|
AS | Assignment |
Owner name: LSI CORPORATION, CALIFORNIA Free format text: TERMINATION AND RELEASE OF SECURITY INTEREST IN PATENT RIGHTS (RELEASES RF 032856-0031);ASSIGNOR:DEUTSCHE BANK AG NEW YORK BRANCH, AS COLLATERAL AGENT;REEL/FRAME:037684/0039 Effective date: 20160201 Owner name: AGERE SYSTEMS LLC, PENNSYLVANIA Free format text: TERMINATION AND RELEASE OF SECURITY INTEREST IN PATENT RIGHTS (RELEASES RF 032856-0031);ASSIGNOR:DEUTSCHE BANK AG NEW YORK BRANCH, AS COLLATERAL AGENT;REEL/FRAME:037684/0039 Effective date: 20160201 |
|
AS | Assignment |
Owner name: BANK OF AMERICA, N.A., AS COLLATERAL AGENT, NORTH CAROLINA Free format text: PATENT SECURITY AGREEMENT;ASSIGNOR:AVAGO TECHNOLOGIES GENERAL IP (SINGAPORE) PTE. LTD.;REEL/FRAME:037808/0001 Effective date: 20160201 Owner name: BANK OF AMERICA, N.A., AS COLLATERAL AGENT, NORTH Free format text: PATENT SECURITY AGREEMENT;ASSIGNOR:AVAGO TECHNOLOGIES GENERAL IP (SINGAPORE) PTE. LTD.;REEL/FRAME:037808/0001 Effective date: 20160201 |
|
AS | Assignment |
Owner name: AVAGO TECHNOLOGIES GENERAL IP (SINGAPORE) PTE. LTD., SINGAPORE Free format text: TERMINATION AND RELEASE OF SECURITY INTEREST IN PATENTS;ASSIGNOR:BANK OF AMERICA, N.A., AS COLLATERAL AGENT;REEL/FRAME:041710/0001 Effective date: 20170119 Owner name: AVAGO TECHNOLOGIES GENERAL IP (SINGAPORE) PTE. LTD Free format text: TERMINATION AND RELEASE OF SECURITY INTEREST IN PATENTS;ASSIGNOR:BANK OF AMERICA, N.A., AS COLLATERAL AGENT;REEL/FRAME:041710/0001 Effective date: 20170119 |
|
AS | Assignment |
Owner name: BELL SEMICONDUCTOR, LLC, ILLINOIS Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:AVAGO TECHNOLOGIES GENERAL IP (SINGAPORE) PTE. LTD.;BROADCOM CORPORATION;REEL/FRAME:044887/0109 Effective date: 20171208 |
|
AS | Assignment |
Owner name: CORTLAND CAPITAL MARKET SERVICES LLC, AS COLLATERA Free format text: SECURITY INTEREST;ASSIGNORS:HILCO PATENT ACQUISITION 56, LLC;BELL SEMICONDUCTOR, LLC;BELL NORTHERN RESEARCH, LLC;REEL/FRAME:045216/0020 Effective date: 20180124 |
|
MAFP | Maintenance fee payment |
Free format text: PAYMENT OF MAINTENANCE FEE, 12TH YEAR, LARGE ENTITY (ORIGINAL EVENT CODE: M1553); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY Year of fee payment: 12 |
|
AS | Assignment |
Owner name: BELL NORTHERN RESEARCH, LLC, ILLINOIS Free format text: RELEASE BY SECURED PARTY;ASSIGNOR:CORTLAND CAPITAL MARKET SERVICES LLC;REEL/FRAME:059720/0223 Effective date: 20220401 Owner name: BELL SEMICONDUCTOR, LLC, ILLINOIS Free format text: RELEASE BY SECURED PARTY;ASSIGNOR:CORTLAND CAPITAL MARKET SERVICES LLC;REEL/FRAME:059720/0223 Effective date: 20220401 Owner name: HILCO PATENT ACQUISITION 56, LLC, ILLINOIS Free format text: RELEASE BY SECURED PARTY;ASSIGNOR:CORTLAND CAPITAL MARKET SERVICES LLC;REEL/FRAME:059720/0223 Effective date: 20220401 |